Journal of Crystal Growth 430 (2015) 6370

Contents lists available at ScienceDirect

Journal of Crystal Growth

journal homepage: www.elsevier.com/locate/jcrysgro

Numerical examination of the effect of steady crucible rotation

in the liquid phase diffusion growth of SiGe
M. Sekhon n, S. Dost
Crystal Growth Laboratory, University of Victoria, Victoria, BC, Canada V8W3P6

art ic l e i nf o a b s t r a c t

Article history: A numerical simulation study has been carried out to investigate the inuence of steady crucible rotation
Received 27 March 2015 on the LPD growth of SixGe1
x. Results reveal the effectiveness of suppressing natural convection driven
Received in revised form ows with the application of ampoule rotation. Improvement in the uniformity of radial composition of
19 August 2015
the growing crystal is predicted after two hours of growth for the remaining simulated growth period.
Accepted 24 August 2015
Further, a marginal reduction in the growth interface curvature due to crucible rotation is also observed
Dr. J.J. Derby
Available online 2 September 2015 in the results. Numerical computation shows the potential of crucible rotation as a viable alternative to
static magnetic eld in inhibiting buoyancy induced ows in the LPD growth of SixGe1
x.
Keywords: & 2015 Elsevier B.V. All rights reserved.
A2. Liquid phase diffusion
A1. Computer simulation
B1. Germanium silicon alloys
A2. Steady crucible rotation

1. Introduction
SixGe1
x is an alloy semiconductor that is becoming increas-
ingly important for various applications primarily due to the
complete solid and liquid state miscibility exhibited by this system
[1]. However, an inherent difculty with this alloy system is the
wide gap between solidus and liquidus lines in its phase diagram
which makes the use of established melt growth techniques dif-
cult for growing uniform composition single crystals. Liquid
phase diffusion (originally developed as a variant of multi-com-
ponent zone melting in [2]) is a solution growth technique that
has been successfully used to grow single crystals of this material
[3]. The LPD experimental set up is comprised of a single crystal
germanium seed, polycrystalline germanium solvent, and poly-
crystalline silicon source placed in a quartz crucible subjected to
an axial temperature gradient as shown in Fig. 1. Due to the
imposed temperature gradient, the solvent material melts com-
pletely, forming the growth solution, the seed melts partially and
the silicon source remains in the solid state due to its higher
melting point. The partially melted seed and molten solvent form
the initial growth interface while the molten solvent and solid
silicon form the dissolution interface. As more and more silicon
dissolves into the solution, eventually the solution near the growth
interface saturates and growth commences in the upward direc-
tion because of the applied axial temperature gradient. During
n
Corresponding author.
E-mail address: mssekhon@uvic.ca (M. Sekhon).
growth, silicon is preferentially consumed from the melt into the
crystal because the solid state solubility of silicon in germanium is
higher than its liquid state value and is replenished by a con-
tinuous supply of silicon from the source at the top. As growth
proceeds in the upward direction, the growth interface tempera-
ture increases and results in an axially graded composition crystal
(details of the experimental process can found in [3]).
It was numerically predicted in [4] that convection is quite
strong in the early hours of growth after which the transport
process becomes diffusion dominated. In the numerical work [5] it
was shown that a static magnetic eld was effective in suppressing
the natural convection. This was followed by the experimental
investigation [6] of the effect of a static magnetic eld on this
growth system in which it was found that the growth rate gets
signicantly reduced in the presence of a static magnetic eld
which in turn was attributed to the change in the thermal eld.
A numerical study was carried out in [7] to investigate the
effect of ampoule rotation on melt ow and dopant segregation in
the vertical Bridgman growth of gallium-doped germanium. It was
found that ampoule rotation had a similar effect as that of an axial
magnetic eld on this growth system and when the convection
was not very strong in the system it could be completely sup-
pressed [7]. On the downside, it was noted that it could lead to
larger radial dopant segregation in the case when convection was
not fully suppressed [7]. In another numerical work [8] in this
direction, it was concluded that steady ampoule rotation worsened
the radial segregation with a marginal improvement in the axial
segregation in the vertical Bridgman growth of cadmium zinc
telluride. The impact of crucible rotation incorporating the effects

http://dx.doi.org/10.1016/j.jcrysgro.2015.08.022
0022-0248/& 2015 Elsevier B.V. All rights reserved.
64 M. Sekhon, S. Dost / Journal of Crystal Growth 430 (2015) 6370
Fig. 1. Schematic description of the LPD system used. The growth interface (the Ge seedGe melt interface) is initially at the melting temperature of Ge. The computational
domain consists of the Si source, SiGe melt (initially Ge), Ge seed, and the vertical wall of the quartz ampoule.
of centrifugal force using a rotating frame of reference was carried
out in [9]. It was found that the ow damping effect of ampoule
rotation can become more effective by the inclusion of centrifugal
force into the numerical analysis [9]. The contribution of ampoule
rotation becomes increasingly important as rotational speed
increases and radial segregation can be reversed after the ow
transition [9]. The use of steady crucible rotation was found to be
effective in suppressing natural convection and attening the
growth interface in liquid encapsulated vertical gradient freeze
growth of InP [10]. In [11], the uniformity in radial composition
was found to improve due to steady crucible rotation in the tra-
veling heater method growth of silicon germanium. The effect of
constant ampoule rotation in vertical Bridgman growth of multi-
crystalline silicon was studied numerically in [12] and radial seg-
regation was found to deteriorate with crucible rotation.
In this work, we have carried out numerical simulations of the
liquid phase diffusion growth subjected to steady crucible rotation
at 15, 20, and 25 rpm with the objective of studying its effect on
the natural convection driven ow and radial composition varia-
tion in the growing crystal. The results show the effectiveness of
ampoule rotation in suppressing buoyant ow, which is otherwise
strong during early hours of growth. Compositional uniformity in
the radial direction is found to improve after the rst two hours of
growth for the remaining simulated growth period with the
application of crucible rotation. Furthermore, interface curvature is
reduced by the usage of ampoule rotation.
2. Numerical simulation
LPD like most crystal growth techniques involves a moving
boundary which can be handled computationally using a xed
grid or a moving grid approach. In this work, we have utilized the
dynamic grid approach implemented in Ansys Fluent by devel-
oping various user-dened functions to move the mesh nodes.
Details pertaining to the implementation of moving mesh simu-
lation in Ansys Fluent are explained in Section 2.3.
The eld equations in this approach involve additional terms
which account for grid motion and are in turn computed from the
space conservation law [13]. It has been shown in the literature
[13] that if the space conservation law is not satised then it can
lead to creation of articial mass sources which can put additional
constraints on time step size if these errors are to be made neg-
ligible. However, in the present simulation, the problem of arti-
cial mass source does not arise as Ansys Fluent makes use of the
space conservation law for computing grid ux rather than com-
puting it directly from grid velocity [14].
The location of the initial growth interface was identied with
the help of an isotherm corresponding to the melting point of
germanium. The mesh is made up of 4674, 1518, 656, 1002 quad-
rilateral elements in the melt, seed, source, and quartz regions
respectively. To handle the pressure velocity coupling the PISO
algorithm [15] was utilized. A time step of 0.1 s was employed to
carry out the simulation.
The numerical solution domain, as shown in Fig. 1, is comprised
of the silicon source at the top, the SiGe melt in the middle, the
Ge-seed below, and the vertical wall of the quartz ampoule. It
should be noted that some of the results presented in this work
which were not directly accessible through the GUI of Ansys Fluent
were obtained by creating appropriate user dened functions.
2.1. Assumptions
For carrying out the numerical simulations a number of sim-
plifying assumptions were made and are summarized as follows:
 Local thermodynamic equilibrium is assumed to exist at the
growth interface.
 The SiGe melt is assumed to be a Newtonian and incompres-
sible uid.
 The melt density change is assumed to be small, and is
approximated in the momentum equation by using the well-
known Boussinesq approximation in terms of the thermal and
solutal expansion coefcients.
 Dissolution interface is assumed to be xed because of the
extremely small silicon dissolution rate observed in the
experiments (around 0.3 mm/day) [4]. Silicon concentration is
approximated using equilibrium composition at this assumed
xed dissolution interface instead of using the more accurate
ux boundary condition. However, the authors believe that this
assumption should not signicantly affect the key ndings of
this comparative numerical simulation study.
 Latent heat of solidication is neglected because of small
growth rate.
 M. Sekhon, S. Dost / Journal of Crystal Growth 430 (2015) 6370 65

 The enthalpy of dissolution and the volume change associated Mass transport
with the dissolution process are neglected as silicon and ger-
d
manium form a nearly ideal mixture [16]. CdV + C n( u ug ) dS = D (nC ) dS
dt
 Furnace heat transfer computation is not included in the V (t ) S (t ) S (t ) (4)
numerical model: instead, the temperature is prescribed along
the crucible wall. Although this represents an idealized
2.2.2. Solid phases
boundary condition the same boundary condition is used with
In the solid phases (namely, grown SiGe crystal, silicon source,
and without crucible rotation so it does not affect the key
and quartz ampoule wall) of the computational domain, the only
ndings of the present work.
 eld equation is the energy balance. Mass transport in the grown
Soret and Dufour effects were not included in the numerical
crystal was not taken into account.
model.
 In the grown crystal, the energy balance reads as follows:
Heat and mass uxes are given by the well-known Fourier's and
Fick's laws, respectively. d
TdV T nug dS = (nT ) dS
 Variation of transport parameters with temperature during dt
V (t ) S (t ) S (t ) (5)
growth is neglected.
 Mass diffusivity of silicon in the grown crystal is neglected in For the domains of the quartz wall and silicon solid source,
comparison to its value in the SiGe (initially Ge) melt. which do not have any moving boundaries, the energy balance
 The growth system is considered axisymmetric, and so are the takes the following form:
eld equations of the liquid and solid phases.
 d
Crucible rotation is initiated after partial melting of the seed is
dt
TdV = (nT ) dS
complete. V S (6)

As pointed out earlier, to determine the grid ux (which is

2.2. Field equations required in discretized eld equations for the moving solid and
uid sub domains), the space conservation law [13] is used and is
The eld equations are derived from the well-known balance given by:
laws of conservation of mass, momentum balance, energy balance, d
and mass transport. dt
dV ( nug ) dS = 0
V (t ) S (t ) (7)

2.2.1. Liquid phase In the discretized form the above equation reads [14]:
The liquid phase is the SiGe solution/melt (initially Ge melt). nf
V n+ 1 V n
The balance laws yield the following eld equations in the liquid = ug, j nj Sj
t j (8)
phase which are respectively known as Continuity, Momentum
Balance, Energy Balance and Mass Transport equations. These Thus the grid ux is computed as the volume swept by face
equations are written for a rotating frame of reference due to j over time step t [14].
which the centrifugal and Coriolis forces appear in the momentum In the above equations, is the density of the domain, n is the
balance. Moreover, the simulations with crucible rotation were exterior unit normal to the closed surface S enclosing volume V,
carried out using the model axisymmetric with swirl in Ansys u and ug represent the ow and grid velocities respectively, m is
Fluent, which essentially means that the tangential component of the dynamic viscosity, p is the pressure, T and C denote the
the velocity is non-zero although the gradients in the tangential thermal and solutal expansion coefcients respectively, g stands
for the gravity vector, is the angular velocity, r represents the
direction are zero [17]. Furthermore, the eld equations are pre-
radial position vector, T denotes temperature while C is the silicon
sented in the integral form for an arbitrary volume V(t) bounded
mass concentration in the melt, and Tref and Cref represent the
by a moving surface S(t) in the moving sub domains (crystal and
reference temperature (1211.87 K) and reference concentration,
melt) and an arbitrary volume V bounded by a surface S in the respectively. and D are the thermal and mass diffusivities of the
xed sub domains (source and quartz). melt, respectively. Superscript n and n 1 represent the current
and next time level, nf denotes the number of faces of a control
Continuity volume and t is the time step size.

(nu) dS = 0 2.3. Dynamic mesh motion implementation in Ansys Fluent

S (t ) (1)
Eq. (18) which is obtained from the mass balance at the growth
Momentum balance
interface assuming equilibrium composition and neglecting mass
d
n( u u g ) udS =
diffusion in the growing crystal is used to compute the growth
dt
udV + [n(u)] dS
V (t ) S (t ) S (t )
interface velocity. This is done by computing the concentration
gradient (with the help of data access macro [18]), liquidus and
pdV g T (T Tref ) dV gC ( C C ref ) dV
solidus equilibrium composition along the growth interface from
V (t ) V (t ) V (t )
Eqs. (16) and (17) respectively. The position of each mesh node
[2 u + ( r)] dV lying on the growth interface is updated during each time step
V (t ) (2)
based on the computed growth interface velocity. To implement
Energy balance this in Ansys Fluent, user dened functions of the type DEFI-
NE_GRID_MOTION [18] were created. In response to the pre-
d scribed boundary motion, the internal mesh in the solid and liquid
TdV + T n( u ug ) dS = (nT ) dS sub domains needs to be adjusted in order to preserve the mesh
dt
V (t ) S (t ) S (t ) (3) quality. As growth proceeds, the size of the liquid sub domain
66 M. Sekhon, S. Dost / Journal of Crystal Growth 430 (2015) 6370
decreases while that of the solid sub domain increases and this
changing size of the liquid and solid sub domains is tackled by the
dynamic layering method [17] of Ansys Fluent as it can be used for
quadrilateral elements in a mesh. As per this method, the cell layer
size in the solid sub domain is allowed to increase until the fol-
lowing condition is met [17]:
hmin > ( 1 + S ) hideal
where hmin is the minimum cell height in the cell layer adjacent to
the boundary, hideal is the ideal cell height and S is the split factor.
Once the above condition is satised, the cell layer is split using
either height based (cell layer is divided to create new cells of
height hideal and h
hideal) or ratio based option (cell layer is
divided in such a way that the ratio of the height of the new cells is
S) [17]. Similarly, in the liquid sub domain the cell layer is com-
pressed until condition given by Eq. (10) is met after which the cell
layer is merged with adjacent cell layer [17].
hmin < C hideal
where C is collapse factor. For the present simulations, ratio based
option was used to split the cell layer, a value of 0.1 was used for
both S and C, and the value of hideal was varied during the course
of simulation in order to maintain a good quality mesh.
In addition to the changing size of melt and crystal sub
domains the growth interface changes in shape as the simulation
proceeds and the internal mesh must adapt itself to this changing
shape so as to maintain the quality of the mesh. Spring based
smoothing was utilized for this purpose in which edges between
two nodes are modeled as network of interconnected springs [17].
In response to the prescribed boundary node motion, the force on
a mesh node can be computed from Hooke's law and is given as
follows [17]:
ni
Fi = kij ( x j x i)
j (11)
where xi and xj are the node displacements of the ith node and
its neighboring jth node, ni is the number of nodes in the neigh-
borhood of ith node which are linked to node i and kij is spring
constant between the nodes i and j and is computed from the
following relation [17]:
kfac
kij =
|x i x j |
where kfac is the spring constant factor. This factor can be used to
manipulate the spring stiffness and it can be specied in between
0 and 1 [17]. A value of 1 means the boundary node displacement
will have lesser inuence on the displacement of interior nodes in
comparison to a value of 0 [17] and was used in the present
simulation.
Further noting that the resultant force on any given node due to
all the springs attached to that node is zero at equilibrium which
leads to the following iterative equation [17]:
Table 1
Coefcients used to compute equilibrium compositions.
Coefcient aL bL
Value 5.72  10
2 1.49  10
4
Coefcient aS bS
Value 3.82  10
1
6.25  10
4
n
j i kij xmj
x im + 1 = n
j i kij (13)
where superscript m indicates the mth iteration.
From the known boundary node displacements the above
iterative equation is solved for the interior mesh nodes until the
solution is converged, after which the node positions are updated
(9) as follows [17]:
x in + 1 = x in + x iconverged (14)
For the present simulation, a convergence tolerance of 10
5
was used.
2.4. Boundary and initial conditions
2.4.1. Concentration eld
The wall of the quartz ampoule is considered impermeable, and
hence the mass balance yields the condition of zero-normal con-
(10) centration gradient on this boundary:
C
=0
n (15)
On the growth interface, it was assumed that local thermo-
dynamic equilibrium exists and the liquid equilibrium concentra-
tion was expressed as a function of temperature using the phase
diagram of the SiGe system. As mentioned earlier, the dissolution
interface was assumed to be xed and silicon concentration was
approximated using equilibrium composition instead of using the
ux boundary condition. Following polynomial expression was
used to compute the liquidus equilibrium composition:
L
Ceq = a LT + b L T2 + c LT 3 + d LT 4 + e L T 5 (16)
where we have calculated T at the growth interface by
T = Tgrowth 938.72, and at the dissolution interface by
T = Tdis 938. 72. The coefcients in Eq. (16) were determined by
curve tting from the SiGe phase diagram for equilibrium (values
are given in Table 1). The solid equilibrium concentration at the
growth interface was also determined in a similar fashion by curve
tting the solidus curve in the SiGe phase diagram by using:
S
Ceq = a S T + b S T2 + c S T 3 + d S T 4 + e S T 5 (17)
where superscripts L and S stand for the liquid and solid phases,
respectively. The subscripts growth and dis represent the
values of the parameters at the growth and dissolution interfaces,
(12) respectively. The coefcients in Eq. (17) were estimated by curve
tting from the SiGe phase diagram for equilibrium (values are
given in Table 1). These equations give the equilibrium composi-
tion in atomic percentage while the computation was carried out
in mass fraction.
Initially, the concentration eld is set to zero as the melt is pure
germanium in the beginning. At the dissolution interface, the
concentration is computed from Eq. (16) corresponding to the
temperature obtained by solving the steady state heat equation.
In order to move the mesh, the growth interface velocity was
computed from the following mass balance equation [4]:
cL dL eL
7.48  10
7
1.84  10
9 2.21  10
12
cS dS eS
5.95  10
7
2.93  10
10 4.45  10
13
 M. Sekhon, S. Dost / Journal of Crystal Growth 430 (2015) 6370 67

C gradient. It is important to point out that LPD represents axially

(
S ug Cgrowth,eq
S L
)
Cgrowth,eq = L D
n (18) stable natural convection system, which essentially means that the
lighter uid is at the top and heavier uid is at the bottom as the
2.4.2. Thermal eld axial density gradient is created due to the applied temperature
At the bottom of the ampoule:
T = 1146.87 (19)

Along the outer wall of the quartz ampoule adjoining the seed and
melt region:
T = 1146.87 + 4000y (20)

where y is the distance between the face centroid and bottom of

the domain.
Along the outer wall adjoining the source region:
T = 1326.87 (21)

The top of the computational domain (above the source) is

open to vacuum, and there is heat loss through this boundary. The
radiative heat loss that occurs along this boundary was taken as

(
kT = T 4 T 4amb ) (22)

Tamb = 1321.87 (23)

In the above equations, k is the thermal conductivity, denotes

the emissivity, s is the StefanBoltzmann constant, and Tamb
Fig. 2. Contours of stream function (kg/s) after one hour of growth time (a) 0 rpm
represents the temperature in the ambient (in K).
(stream function interval max/50, where max 1.61  10
4 kg/s) (b) 20 rpm
As the latent heat of solidication was neglected due to slow (stream function interval max/40, where max 8.96  10
6 kg/s).
growth, the energy balance at the growth interface leads to the
condition of continuous heat ux:

T S T L
kS = kL
n n (24)

where k S and k L represent the thermal conductivity of the solid

and liquid phases, respectively. The above condition was also used
at all other inner boundaries.
The initial condition for the temperature eld was obtained by
solving the steady state heat equation.

2.4.3. Flow eld

A no-slip boundary condition was used for the ow velocity
eld along the boundaries of the melt. Initially, the ow velocities
were set to zero.
The values of the physical properties used in the present work
are listed in Table 2 [4,1921].

3. Results and discussion

Firstly, the effect of crucible rotation on natural convection is

examined. As can be seen from Figs. 2a and 3a natural convection
Fig. 3. Contours of stream function (kg/s) after two hour of growth time at
is quite strong during the early hours of growth without any (a) 0 rpm (stream function interval max/50, where max 1.26  10
4 kg/s)
ampoule rotation and can be attributed to the radial temperature (b) 20 rpm (stream function interval max/40, where max 2.84  10
6 kg/s).

Table 2
Thermo-physical properties of the silicongermanium system.

Property Symbol (Units) Source Seed Solution/Melt Crystal Quartz

Thermal conductivity k (W/m K) 23.7 10.6 42.8 10.6 2.0

Density (kg/m3) 2301.6 5323.0 5633.0 5323.0 2200.0
Thermal diffusivity (m2/s) 1.06  10
5 5.03  10
6 1.87  10
5 4.09  10
6 7.58  10
7
Mass diffusivity D (m2/s) 10
8
Emissivity 0.71 0.85
Thermal expansion coefcient T (1/K) 1.2  10
4
Solutal expansion coefcient C (1/wt%) 0.012
68 M. Sekhon, S. Dost / Journal of Crystal Growth 430 (2015) 6370
and concentration gradients. However, in the beginning the sta-
bilizing effect due to the imposed concentration gradient is neg-
ligible as not enough silicon has dissolved into the melt and con-
sequently natural convection is strong during early growth hours
[4]. It is clear from Figs. 2b and 3b that with the application of
crucible rotation, the buoyancy induced ow is signicantly sup-
pressed near the growth interface. A sharp decrease in axial and
radial components of the velocity with the application of crucible
rotation can be readily observed in Fig. 4. The ow damping effect
of crucible rotation is also evident from Fig. 5 which shows that
the swirl velocity contours have straightened considerably for the
examined range of rotational speeds. Furthermore, as expected
mass transport is slowed down with crucible rotation during early
growth hours due to suppression of convection as can be observed
from Fig. 6.
The radial composition of the simulated crystal growth for
various crucible rotation rates at different growth times is depicted
in Fig. 7. As can be seen from Fig. 7a with the application of cru-
cible rotation, there is not much gain in terms of uniformity in
radial composition till two hours of growth, in fact there is a slight
increase in the non-uniformity during this period. This can be
attributed to the fact that the ampoule rotation suppresses the
natural convective ow which results in poor mixing in the radial
direction and in turn leads to slightly larger compositional non-
uniformity in the radial direction. After two hours of growth,
Fig. 4. Maximum absolute axial and radial velocity components near growth
interface at various rotation rates.
Fig. 5. Contours of swirl velocity (m/s) after 1 h of growth time (swirl velocity contour interval v,max/20) at (a) 15 rpm (v,max 1.96  10
2), (b) 20 rpm
(v,max 2.62  10
2) and (c) 25 rpm (v, max 3.27  10
2).
ampoule rotation starts improving the uniformity in the radial
composition which is due to the fact that the buoyancy induced
ow is strong only during the early growth hours and conse-
quently the ow inhibiting effect of crucible rotation is absent in
later hours of growth. As the growth proceeds, the interface cur-
vature increases (see Fig. 8a) and consequently an upward trend in
the silicon composition in the radially outward direction is
observed for the remaining simulated growth period. As growth
continues, ampoule rotation reduces the non-uniformity in the
radial composition but is not able to achieve the desired outcome
of completely uniform radial composition. The reason for this can
be understood from Fig. 8bd which shows that the usage of
steady ampoule rotation does not completely eliminate the inter-
face curvature. As a result, an upward trend in the radial compo-
sition is observed in Fig. 7be for all rotational speeds considered.
It can also be inferred from Fig. 7 that increasing the rotational
speed does not lead to any further signicant improvement in the
Fig. 6. Contours of concentration (% atm) after 1.5 hour of growth time (con-
centration contour interval (cmax -cmin)/20) at a) 0 rpm (cmax 5.47, cmin 0.037)
b) 20 rpm (cmax 5.47, cmin 0.002).
 M. Sekhon, S. Dost / Journal of Crystal Growth 430 (2015) 6370 69

Fig. 7. Radial composition plots for various rotational rates at different growth times.

Fig. 8. Interface evolution plot from t 3 h to t 19 h with a time interval of four hours at (a) 0 rpm, (b) 15 rpm (c) 20 rpm and (d) 25 rpm.
70 M. Sekhon, S. Dost / Journal of Crystal Growth 430 (2015) 6370
Fig. 9. Growth velocity prole along the centreline for various rotational rates.
Fig. 10. Axial silicon composition along the centreline for various rotational rates.
uniformity of radial composition. The growth velocity prole for
various considered cases is shown in Fig. 9. Without crucible
rotation, a signicant dip in the growth velocity is observed after
the early growth hours which can be attributed to the weakening
of convection as the growth proceeds due to stabilizing axial
density gradient. After this, the growth velocity recovers and
eventually starts to decrease again. With the application of cruci-
ble rotation, growth velocity is signicantly reduced due to sup-
pression of convection during the early growth hours. This is also
evident from Fig. 7a and b which clearly shows that crystal com-
position with crucible rotation lags behind the case without any
rotation during early growth hours. After the initial growth period,
the growth velocity along the centreline with crucible rotation
increases in comparison to the case without rotation and helps in
attening the growth interface (as can be seen in Fig. 8bd) and
improving the uniformity in radial composition in later hours of
growth. Finally, it can be observed from Fig. 10 that the crucible
rotation does not signicantly alter the axial composition
distribution.
4. Conclusion
The effect of steady crucible rotation on the LPD growth system
was examined numerically. Results indicate that buoyancy
induced ow could be signicantly suppressed with the help of
ampoule rotation. After rst two hours of growth, uniformity in
the radial composition was found to improve with the usage of
crucible rotation for the remaining simulated growth period.
However, complete uniformity in radial composition could not be
accomplished. Improvement in the radial compositional uni-
formity was found to be insensitive to the investigated range of
rotational speeds. Furthermore, simulations also revealed that
ampoule rotation helps in reducing interface curvature but a
complete interface attening effect could not be achieved. The
promising computational results obtained in this work, make a
strong case for employing crucible rotation for damping natural
convection driven ow instead of axial static magnetic eld which
was found to hamper the growth rate signicantly [6]. However,
these ndings must be validated experimentally before a nal
conclusion can be drawn regarding suitability of the usage of
crucible rotation in the LPD growth and a value between 15 and
20 rpm seems to be appropriate to carry out the experiments.
Acknowledgment
Authors express their deepest appreciation to Mr. Brian Lent for
his critical reading of the manuscript and his invaluable sugges-
tions. Financial support provided by the Natural Sciences and
Engineering Research Council of Canada (NSERC) and the Canada
Research Chairs (CRC) Program is gratefully acknowledged.
References
[1] E. Kasper, J. Cryst. Growth 150 (1995) 921.
[2] K. Nakajima, S. Kodama, S. Miyashita, G. Sazaki, S. Hiyamizu, J. Cryst. Growth
205 (1999) 270.
[3] M. Yildiz, S. Dost, B. Lent, J. Cryst.Growth 280 (2005) 151.
[4] M. Yildiz, S. Dost, Int. J. Eng. Sci. 43 (2005) 1059.
[5] E. Yildiz, S. Dost, M. Yildiz, J. Cryst. Growth 291 (2006) 497.
[6] N. Armour, S. Dost, Cryst. Res. Technol. 45 (2010) 244.
[7] C.W. Lan, J. Cryst. Growth 197 (1999) 983.
[8] A. Yeckel, F.P. Doty, J.J. Derby, J. Cryst. Growth 203 (1999) 87.
[9] C.W. Lan, J. Cryst. Growth 229 (2001) 595.
[10] Y. Okano, H. Kondo, S. Dost, J. Cryst. Growth 237239 (2002) 1769.
[11] Y. Takagi, Y. Okano, S. Dost, J. Heat Transf. 134 (2012) 1.
[12] M.P. Bellmann, E.A. Meese, J. Cryst. Growth 333 (2011) 1.
[13] I. Demirdzic, M. Peric, Int. J. Numer. Methods Fluids 8 (1988) 1037.
[14] Ansys Fluent 14.5 Theory Guide, Ansys Inc., Canonsburg, PA, 2012.
[15] R.I. Issa, J. Comput. Phys. 62 (1986) 40.
[16] R.G. Mortimer, Physical Chemistry, third ed, Academic press, San Diego, CA,
2008.
[17] Ansys Fluent 14.5 User's Guide, Ansys Inc., Canonsburg, PA, 2012.
[18] Ansys Fluent 14.5 UDF Manual, Ansys Inc., Canonsburg, PA, 2012.
[19] A. Virzi, J. Cryst. Growth 112 (1991) 699.
[20] S. Yesilyurt, L. Vujisic, S. Motakef, F.R. Szofran, M.P. Volz, J. Cryst. Growth
207 (1999) 278.
[21] A. Rouzaud, D. Camel, J.J. Favier, J. Cryst. Growth 73 (1985) 149.